N-29 Alpha Matte Extractor
JHPixelProAlphaMatteExtractor extracts a soft alpha matte from the pack's strict 3-value trimap convention using the Levin 2008 closed-form matting Laplacian. In practice, it is the quality-first mask-finishing node for hair, fur, and semi-transparent edges where binary masks are not good enough. The current pack version adds CUDA acceleration through PyTorch sparse solvers while preserving an exact CPU fallback.
Schema
| Name | Kind | Type / default | Description |
|---|---|---|---|
trimap |
Input | MASK |
3-value trimap with 0.0 background, 0.5 unknown, 1.0 foreground and ±0.05 tolerance. |
guide |
Input | IMAGE |
Guide RGB image used to build the local color covariance model. |
epsilon |
Widget | FLOAT, default 0.0000001 |
Numerical regularizer for the local covariance estimate. |
window_radius |
Widget | INT, default 1 |
Local window radius for the Levin matting neighborhood. |
lambda_constraint |
Widget | FLOAT, default 100.0 |
Constraint weight that ties the solve back to known trimap regions. |
compute_device |
Widget | COMBO, default auto |
Picks auto, cuda, or cpu for the sparse solve path. |
alpha |
Output | MASK |
Final soft alpha matte for downstream compositing or preview. |
Workflow preview
Workflow JSON: workflows/N-29-alpha-matte-extractor.json